RNA sequencing identifies gene regulatory networks controlling extracellular matrix synthesis in intervertebral disk tissues

Abstract

Degenerative disk disease of the spine is a major cause of back pain and disability. Optimization of regenerative medical therapies for degenerative disk disease requires a deep mechanistic understanding of the factors controlling the structural integrity of spinal tissues. In this investigation, we sought to identify candidate regulatory genes controlling extracellular matrix synthesis in spinal tissues. To achieve this goal we performed high throughput next generation RNA sequencing on 39 annulus fibrosus and 21 nucleus pulposus human tissue samples. Specimens were collected from patients undergoing surgical discectomy for the treatment of degenerative disk disease. Our studies identified associations between extracellular matrix genes, growth factors, and other important regulatory molecules. The fibrous matrix characteristic of annulus fibrosus was associated with expression of the growth factors platelet derived growth factor beta (PDGFB), vascular endothelial growth factor C (VEGFC), and fibroblast growth factor 9 (FGF9). Additionally we observed high expression of multiple signaling proteins involved in the NOTCH and WNT signaling cascades. Nucleus pulposus extracellular matrix related genes were associated with the expression of numerous diffusible growth factors largely associated with the transforming growth signaling cascade, including transforming factor alpha (TGFA), inhibin alpha (INHA), inhibin beta A (INHBA), bone morphogenetic proteins (BMP2, BMP6), and others.

Introduction

Back pain is among the leading global causes of disability^{1, 2}, with degenerative disk disease and osteoarthritis being important causes of disease. Disk degeneration is caused by a dysregulation of extracellular matrix homeostasis, characterized by dehydration of the central nucleus, reduced proteoglycan content, decreased cellularity, diminished endplate density, and disruption of the annulus^3–5. Environmental exposures, as well as genetic and epigenetic factors have been associated with disk degeneration and altered extracellular matrix synthesis in disk tissues^6,7. Novel molecular approaches that can target molecular factors regulating extracellular matrix synthesis in disk tissue have the potential to be used as therapeutic agents to slow or reverse disk degeneration in patients.

The molecular phenotype of intervertebral spinal disk tissue, including the annulus fibrosus (AF) and nucleus pulposus (NP), has been studied extensively in non-human animal models for degenerative disk disease^8–14. Studies evaluating transcriptome data using microarrays have provided us with an initial understanding of the molecular mechanisms underlying disk biology and have played a major role in helping to identify important biologic markers specific for AF and NP disk tissues^15–19. However knowledge regarding the regulatory role of molecular factors and how they contribute to tissue homeostasis still requires further study.

In this investigation we seek to identify molecular regulatory factors whose transcriptional profiles correlate with the expression of extracellular matrix proteins important for the structural phenotype of human AF and NP tissues. To achieve this objective we evaluated transcriptome profiles of a cohort of human cervical disk tissue samples utilizing high throughput next generation RNA sequencing. We obtained complete gene expression profiles for 60 surgically harvested cervical disk specimens (AF and NP), and evaluated the main molecular landscapes of these two principal disc tissues. The large cohort of samples analyzed in this study allowed us to successfully perform weighted gene correlation analysis to identify gene regulatory clusters in disk tissues and assess gene relationships.

The molecular regulators that show relationships with extracellular matrix gene expression represent promising candidates for future study and therapeutic validation. The findings in this investigation also serve to support regenerative medicine therapies currently under development for the treatment of intervertebral disk disease, including stem cell therapies and tissue engineering strategies to regrow disk tissue for surgical transplantation and disk replacement procedures^20,21. Both of these strategies require a comprehensive definition of the molecular phenotype of the human intervertebral disk to evaluate the efficacy of strategies to differentiate stem cells or engineer tissue disk tissue in vivo. The transcriptional signatures and gene relationships identified in this study have broad applicability in both the stem cell and tissue engineering fields.

Methods

Surgical tissue collection

A total of 60 tissue specimens were collected for research use from 48 adult patients undergoing cervical discectomy. Patients ranged in age from 32 to 77 years of age and included a balanced distribution of male and female patients (Supplemental Table 1). Patients in this study underwent surgery for the treatment of symptomtic degenerative disk disease presenting with or without myelopathy. Subjects were enrolled in the study in the period between January 2011 and April 2015. Cases in which discectomy was performed in the setting of acute trauma or infection were excluded from this study. At the time of tissue collection, the AF and NP were carefully dissected from one another in the operating room by the staff surgeon. In cases where disc degeneration was severe, NP tissue could not always be readily identified and distinguished from the AF tissue and therefore could not be collected for some patients. At the time of surgical harvest, tissues were snap frozen in liquid nitrogen and stored at −80°C until ready for RNA extraction. All samples were frozen within 40 minutes of removal from the patient. Grade of disk degeneration was evaluated on preoperative lateral radiographs and was characterized using the classification described by Lane et al.^22,23. Clinical data available for each disk sample is provided in Supplemental Table 1. The specimens used in this investigation were collected under institutional review board approved protocols (IRB#10–005713). Written informed consent was obtained for all biospecimens that were analyzed.

RNA extraction from intervertebral disk tissue

Frozen tissue biopsies were ground into a powder using a mortar and pestle and homogenized in Qiazol reagent (Qiagen, Hilden, Germany) and homogenized. Total RNA was extracted from research biopsies based on previous methods^{24, 25} using the miRNeasy minikit (Qiagen, Hilden, Germany) and quantified using the NanoDrop 2000 spectrophotometer (Thermo Fischer Scientific, Wilmington, Delaware). For samples selected for next generation sequencing, RNA integrity was assessed using the Agilent Bioanalyzer DNA 1000 chip (Invitrogen, Carlsbad, CA).

Next generation mRNA sequencing, statistics and bioinformatics

RNA sequencing and bioinformatics analyses were performed as previously described^26–29. In brief, library preparation was performed using the TruSeq RNA library preparation kit (Illumina, San Diego, CA). Polyadenylated mRNAs were selected using oligo dT magnetic beads. TruSeq Kits (12-Set A and 12-Set B) were used for indexing to permit multiplex sample loading on the flow cells. Paired-end sequencing reads were generated on the Illumina HiSeq 2000 sequencer. Quality control for concentration and library size distribution was performed using an Agilent Bioanalyzer DNA 1000 chip and Qubit fluorometry (Invitrogen, Carlsbad, CA). Sequence alignment of reads and determination of normalized gene counts were performed using the MAP-RSeq (v.1.2.1) workflow³⁰, utilizing TopHat 2.0.6^{31, 32}, and HTSeq³³.

RNA sequencing data were analyzed to assess relevant genes that differ between AF and NP specimens. Genes with a minimal expression value (RPKM > 0.01) were included in subsequent computational analysis. Fold-change differences in gene expression were evaluated using the Mann-Whitney U test with a 1% false discovery rate (FDR), and statistical significance was set at p < 0.05. Unsupervised hierarchical clustering was performed using the Pearson correlation method. Weighted gene correlation analysis was performed using the R package WGCNA (Weighted Gene Correlation Analysis)³⁴. Genes with an average RPKM expression > 0.01 across all specimens were included in the computational analysis. Functional gene annotation classification of WGCNA clusters was performed using DAVID Bioinformatics Resources 6.7 database (DAVID 6.7)³⁵.

Results

RNA sequencing was performed using 60 unique cervical spine disk tissue samples (39 AF and 21 NP specimens). High quality sequencing reads were obtained for 57 of the 60 samples. The 3 samples with abnormally low read counts were excluded from further analysis. To detect sample outliers, an unbiased assessment of transcriptome data using unsupervised hierarchical clustering was performed. This analysis revealed 10 disk samples that clustered independently from the majority of the disk specimens (Supplemental Figure 1). A comparison of these samples with specimens in the primary cluster show that outlier samples express higher levels of blood related genes including genes linked to the erythroid, lymphoid, and myeloid lineages. A selective evaluation of the blood specific hemoglobin genes, hemoglobin subunit beta (HBB), hemoglobin subunit alpha 1 (HBA1), and hemoglobin subunit alpha 2 (HBA2), confirmed that these 10 samples express the highest levels of blood related genes (Supplemental Table 2). To ensure the comparability of samples in our study these outliers were removed from subsequent analysis. Repeated unsupervised hierarchical clustering after removal of outliers showed an expected trend toward independent clustering of AF and nucleus puplosus tissue samples (Figure 1a). Genes that are not strictly linked to a tissue phenotype such as hematopoietic and inflammation related genes, as well as tissue heterogeneity, played a role in defining the clustering dendrogram. This observation explains why the clustering dendrogram showed a trending, but not completely independent clustering of the AF and NP specimens, despite the distinct biological phenotypes of AF and NP tissues.

Figure 1.

(a) Unsupervised hierarchical clustering of RNA sequencing data after removal of sample outliers. In this clustering scheme there is a trend for AF and NP samples to preferentially cluster separately. These findings suggest that there are tissue specific differences contained within the transcriptome data, representing the known biological differences that exist between these two tissue types. Our unbiased approach also incorporates various factors that are not directly related to the disc phenotype such as tissue heterogeneity, blood content, and inflammation, which can drive some of the biological variation between specimens, thus precluding a perfect clustering dendogram in which AF and NP specimens cluster as completely independent groups. (b) Genes expressed > 100 RPKM in surgically isolated AF and NP tissue with equal expression levels (Fold change <1.5 between AF and NP). This analysis shows that AF and NP both share common expression of a large number of housekeeping genes as well as a small number of extracellular matrix proteins and growth factor binding associated proteins. (c) Gene ontology analysis reveals enrichment in pathways that promote cellular adhesion including genes linked to notch signaling (vasculature development, GTPase regulator activity) in AF tissue. (d) The NP shows enrichment in genes linked to extracellular matrix protein synthesis, including in genes controlling the extracellular matrix protein synthesis machinery (golgi complex and endoplasmic reticulum).

An examination of the most highly expressed genes (expression > 100 RPKM) commonly expressed in AF and NP expectedly showed common enrichment of genes associated with housekeeping functions (i.e. translation, protein ubiquitination) (Figure 1b). The AF and NP samples share many ECM related genes in common among their highest expressed genes, however the abundance of each gene and their ratios are quite different between AF and NP samples. Of the genes that are commonly enriched in AF and NP that are not associated with house-keeping functions, the AF samples showed higher expression of mRNAs encoding ECM proteins associated with a fibrous matrix including type I collagen (COL1A2), and type VI collagen (COL6A1, COL6A2, COL6A3) (Table 1). In contrast, the NP samples showed increased mRNA levels of genes encoding extracellular matrix proteins associated with a proteoglycan rich chondrogenic matrix, including cartilage oligomeric protein (COMP), lumican (LUM), type II collagen (COL2A1), cartilage intermediate layer protein (CILP), biglycan (BGN), aggrecan (ACAN), type III collagen (COL3A1), chondroadherin (CHAD), and others (Table 2)

Table 1.

Extracellular matrix related genes highly expressed in annulus fibrosus

GeneID	Average expression (RPKM) in annulus fibrosus	GeneID	Average expression (RPKM) in annulus fibrosus
COMP	3990.88	SERPINA1	247.96
FN1	3374.09	TIMP2	246.71
CLU	2562.49	CALR	242.68
TPT1	2502.16	CRTAC1	242.53
DCN	2434.52	DPT	204.77
MGP	2270.18	SERPINF1	198.73
LUM	2039.40	SERPING1	186.09
COL1A1	1393.37	CILP2	185.94
SPARC	1386.58	APOE	179.84
FMOD	1325.29	MMP14	175.88
COL1A2	1255.57	TGFBI	174.91
COL2A1	1161.98	POSTN	173.01
CI LP	1126.00	IBSP	168.53
COL3A1	1042.38	IGFBP4	165.21
CST3	972.51	COL9A3	154.93
HTRA1	871.24	SOD3	152.31
BGN	865.06	IGFBP6	145.98
FGFBP2	860.36	SPARCL1	142.91
LGALS1	711.77	SOD1	137.06
SPP1	681.08	BGLAP	135.65
PRELP	640.12	PLA2G2A	134.42
SCRG1	629.02	APOD	131.80
CHAD	614.98	CHI3L2	130.44
COL6A2	567.11	ANGPTL2	127.09
ACAN	564.31	TIMP3	126.93
CTSK	534.82	FSTL1	126.74
TIMP1	469.09	SERPINE2	125.99
GPX3	450.88	ALDOA	125.16
CTGF	433.32	PRDX4	123.56
MMP9	416.09	CCDC80	121.68
IGFBP7	352.81	COL11A2	117.92
PSAP	349.13	COL5A2	112.32
COL6A1	313.96	NUCB1	111.91
ASPN	308.16	A2M	111.41
MFGE8	292.56	COL6A3	106.22
CYTL1	277.74	LGALS3	105.95
GSN	277.71	FXYD6	104.61
OGN	259.14	ANXA2	100.74

Table 2.

Extracellular matrix related genes highly expressed in nucleus pulposus

GeneID	Average expression (RPKM) in nucleus pulposus	GeneID	Average expression (RPKM) in nucleus pulposus
FN1	6385.17	TIMP2	262.80
COMP	6014.23	CILP2	261.34
CLU	4378.49	CALR	261.25
DCN	3295.38	PLA2G2A	254.91
LUM	3093.90	COL9A3	245.78
MGP	2855.43	TGFBI	242.75
TPT1	2341.70	GSN	242.39
FMOD	2199.02	SERPING1	233.51
COL2A1	1840.27	SERPINE2	226.51
CILP	1626.80	SOD3	201.91
HTRA1	1482.74	IBSP	184.82
FGFBP2	1220.04	CHI3L1	184.64
BGN	1155.23	COL11A2	181.05
SPARC	1065.99	TIMP3	180.53
ACAN	1030.66	CCDC80	175.94
SCRG1	1028.47	COL9A2	168.77
PRELP	989.54	IGFBP6	160.61
COL3A1	940.80	POSTN	158.17
CHAD	835.31	SOD1	155.23
GPX3	643.54	FSTL1	153.32
CTGF	567.45	SPP1	152.31
CHI3L2	541.51	PRDX4	150.92
COL1A2	534.43	FXYD6	147.34
TIMP1	529.60	ANGPTL2	145.53
LGALS1	516.12	IGFBP4	138.17
CRTAC1	475.35	RBP4	137.83
CYTL1	452.88	NUCB1	124.30
CST3	437.51	ALDOA	122.41
COL6A2	430.48	APOE	121.20
SERPINA1	411.25	COL11A1	118.90
OGN	411.25	LGALS3	118.41
PSAP	402.65	APOD	116.58
MFGE8	337.14	COL5A2	116.18
COL1A1	322.44	COL6A3	111.84
ASPN	314.72	CTSK	111.52
DPT	306.39	CRLF1	110.49
COL6A1	285.52	ANXA2	103.52
IGFBP7	278.01	MIA	103.03

To determine genes that are differentially expressed between AF and NP tissues irrespective of their overall abundance, a fold-change comparison of gene expression data in AF and NP was performed. We observed statistically significant enrichment of 1399 genes in AF tissue and 373 genes with enrichment in NP tissue (Supplemental Table 3). Analysis revealed differential gene expression consistent with the biological properties and function of each tissue type. The AF showed enrichment in genes linked to adhesion and regulation of cell contact, consistent with its fibrous structural properties (Figure 1c). In contrast, the NP samples showed enrichment in mRNAs associated with proteoglycan extracellular matrix synthesis, including genes associated with the endoplasmic reticulum and Golgi apparatus (Figure 1d). These findings are consistent with the functional role of the NP, which acts as a hydrostatic cushion to reduce contact pressure between the bony vertebral bodies of the spine. We also observed preferential expression of the notochord specific transcription factor brachyury (T) in NP tissues at low, but detectable levels in about half of the samples. This indicates that residual notochord cell populations, detectable when highly sensitive molecular techniques are applied, may be present in degenerative adult disc tissue.

The AF and NP specimens both showed statistically significant enrichment in known, as well as novel, extracellular matrix proteins and signaling molecules. The AF specimens showed expression of phenotypically important genes such as type IV collagen (COL4A1), multiple laminins important for cell adhesion (LAMA3, LAMA4, LAMA5), and genes linked to NOTCH signaling (DLL1, JAG1, JAG2, NOTCH3, NOTCH4). In NP specimens we observed expression of genes promoting a proteoglycan rich ECM including aggrecan (ACAN), type XI collagen (COL11 A1), glypican 6 (GPC6), lumican (LUM), among others in NP specimens (Table 3).

Table 3.

Significant extracellular matrix related genes enriched in annulus fibrosus and nucleus pulposus

Genes enriched in nucleus pulposus		Genes enriched in nucleus pulposus
CCBE1	collagen and calcium binding EGF domains 1	ACAN	aggrecan
CNTN1	contactin 1	CHI3L1	chitinase 3 like 1
CNTNAP3B	contactin associated protein-like 3B	CHRD	chordin
COL14A1	collagen type XIV alpha 1 chain	COL10A1	collagen type X alpha 1 chain
COL17A1	collagen type XVII alpha 1 chain	COL11A1	collagen type XI alpha 1 chain
COL18A1	collagen type XVIII alpha 1 chain	COL8A2	collagen type VIII alpha 2 chain
COL21A1	collagen type XXI alpha 1 chain	COL9A2	collagen type IX alpha 2 chain
COL24A1	collagen type XXIV alpha 1 chain	CRTAC1	cartilage acidic protein 1
COL4A1	collagen type IV alpha 1 chain	FMOD	fibromodulin
DLL1	delta like canonical Notch ligand 1	FN1	fibronectin 1
DTX1	deltex E3 ubiquitin ligase 1	GPC6	glypican 6
DTX4	deltex E3 ubiquitin ligase 4	HHIPL1	HHIP like 1
EGFLAM	EGF like, fibronectin type III and laminin G domains	HHIPL2	HHIP like 2
JAG1	jagged 1	LAMC3	laminin subunit gamma 3
JAG2	jagged 2	LTBP2	latent transforming growth factor beta binding protein 2
LAMA3	laminin subunit alpha 3	LUM	lumican
LAMA4	laminin subunit alpha 4	OGN	osteoglycin
LAMA5	laminin subunit alpha 5	PRG4	proteoglycan 4
NOTCH3	notch 3	SDC4	syndecan 4
NOTCH4	notch 4	SRPX2	sushi repeat containing protein, X-linked 2
PDGFB	platelet derived growth factor subunit B	WISP3	WNT1 inducible signaling pathway protein 3

Given the heterogeneous nature of spinal tissues, statistical methods used to assess simple fold-change analyses may not always be able to identify all important biological gene relationships. To overcome this challenge and identify novel gene regulatory networks with a functional role in regulating extracellular matrix production, we performed weighted gene correlation network analysis for spine tissues using the R package WGCNA³⁴. Gene correlation analysis identified 46 regulatory gene clusters present in our intervertebral disk samples (Figure 2). We observed gene regulatory clusters associated with housekeeping functions (i.e. translation, transcription, mitochondrion, nuclear homeostasis), cellular infiltration including blood and inflammatory cells. We also observed gene regulatory clusters associated with non-disk tissue including processes related to muscle, bone, and adipogenesis, which likely represent small quantities of tissue mixed in with disk tissue at the time of surgical harvesting.

Figure 2.

Gene correlation networks predicted using weighted genes correlation analysis (WGCNA). Gene networks are associated with a variety of cellular activities including cellular housekeeping, mitosis, tissue heterogeneity, extracellular matrix synthesis as well as numerous others. Gene clusters “paleturquoise”, “darkorange2”, and “darkslateblue” are enriched in known extracellular matrix protein markers in AF, while the clusters “black”, “grey60”, and “lightyellow” are associated with extracellular matrix protein markers characteristic of NP.

To identify novel extracellular matrix proteins and regulatory molecules that control tissue specific phenotypes, we examined clusters containing genes associated extracellular matrix synthesis. The related clusters “paleturquoise”, “darkorange2”, and “darkslateblue” each show enrichment in extracellular matrix proteins and adhesive proteins associated with a fibrous matrix, which is typically characteristic of AF tissue. These clusters contain genes that promote a strong fibrous matrix, including collagens, fibulins, integrins, lamamins, elastin, and others (Table 4). These gene clusters were notably associated with a three diffusible growth factors, fibroblast growth factor 9 (FGF9), platelet-derived growth factor beta polypeptide (PDGFB), and vascular endothelial growth factor C (VEGFC). These findings suggest that these growth factors may play a regulatory roles in maintenance of the AF phenotype and warrant further investigation. Additionally, these clusters also exhibited strong enrichment in genes linked to cell-cell signaling interactions, including the Wnt signaling and NOTCH signaling pathways. Both of these pathways are known to be involved in mediating cell-cell interactions and cellular adhesion in various tissues outside of intervertebral disk^{36, 37}. Given the paucity of diffusible growth factors and the fact that AF cells are in close contact with one another, these data suggest that AF ECM production may be regulated or strongly influenced by direct cell-cell signaling mechanisms, possibly mediated through the Wnt and NOTCH signaling pathways.

Table 4.

Annulus fibrosus co-regulatory gene networks

ECM and cell adhesion related genes				Signaling Associated Genes
Gene symbol	Gene name	Function	Cluster	Gene symbol	Gene name	Function	Cluster
ADAM12	ADAM metallopeptidase domain 12	ECM	"paleturquoise"	FGF9	fibroblast growth factor 9	Growth factor	"paleturquoise"
COL5A1	collagen, type V, alpha 1	ECM	"paleturquoise"	KREMEN1	kringle containing transmembrane protein 1	Wnt signaling	"paleturquoise"
COL5A2	collagen, type V, alpha 2	ECM	"paleturquoise"	PDGFRA	platelet-derived growth factor receptor, alpha polypeptide	Growth factor	"paleturquoise"
COL6A3	collagen, type VI, alpha 3	ECM	"paleturquoise"	WISP1	WNT1 inducible signaling pathway protein 1	Wnt signaling	"paleturquoise"
CDH5	cadherin 5, type 2	Adhesion	"darkorange2"	WNT5A	wingless-type MMTV integration site family, member 5A	Wnt signaling	"paleturquoise"
CDH6	cadherin 6, type 2, K-cadherin	Adhesion	"darkorange2"	WNT6	wingless-type MMTV integration site family, member 6	Wnt signaling	"paleturquoise"
CDH24	cadherin-like 24	Adhesion	"darkorange2"	WNT9B	wingless-type MMTV integration site family, member 9B	Wnt signaling	"paleturquoise"
COL17A1	collagen type XVII alpha 1 chain	ECM	"darkorange2"	CDH6	cadherin 6	NOTCH signaling	"darkorange2"
COL18A1	collagen, type XVIII, alpha 1	ECM	"darkorange2"	CDKN1B	cyclin dependent kinase inhibitor 1B	NOTCH signaling	"darkorange2"
COL21A1	collagen, type XXI, alpha 1	ECM	"darkorange2"	DNER	delta/notch like EGF repeat containing	NOTCH signaling	"darkorange2"
COL4A1	collagen type IV alpha 1 chain	ECM	"darkorange2"	HES5	hes family bHLH transcription factor 5	NOTCH signaling	"darkorange2"
COL4A2	collagen type IV alpha 2 chain	ECM	"darkorange2"	HEYL	hes related family bHLH transcription factor with YRPW motif-like	NOTCH signaling	"darkorange2"
COL4A5	collagen type IV alpha 5 chain	ECM	"darkorange2"	HHEX	hematopoietically expressed homeobox	NOTCH signaling	"darkorange2"
CNTN4	contactin 4	Adhesion	"darkorange2"	HOXD3	homeobox D3	NOTCH signaling	"darkorange2"
ELN	elastin	ECM	"darkorange2"	IGFBP4	insulin-like growth factor binding protein 4	Wnt signaling	"darkorange2"
FBLN1	fibulin 1	ECM	"darkorange2"	IGFBP6	insulin-like growth factor binding protein 6	Wnt signaling	"darkorange2"
FBLN5	fibulin 5	ECM	"darkorange2"	IGFBP7	insulin-like growth factor binding protein 7	Wnt signaling	"darkorange2"
ICAM1	intercellular adhesion molecule 1	Adhesion	"darkorange2"	JAG1	jagged 1	NOTCH signaling	"darkorange2"
ICAM2	intercellular adhesion molecule 2	Adhesion	"darkorange2"	JAG2	jagged 2	NOTCH signaling	"darkorange2"
ITGA3	integrin subunit alpha 3	Adhesion	"darkorange2"	KCNA5	potassium voltage-gated channel subfamily A member 5	NOTCH signaling	"darkorange2"
ITGA6	integrin subunit alpha 6	Adhesion	"darkorange2"	MAML3	mastermind like transcriptional coactivator 3	NOTCH signaling	"darkorange2"
ITGA7	integrin subunit alpha 7	Adhesion	"darkorange2"	NEURL1B	neuralized E3 ubiquitin protein ligase 1B	NOTCH signaling	"darkorange2"
ITGA8	integrin subunit alpha 8	Adhesion	"darkorange2"	NOTCH3	notch 3	NOTCH signaling	"darkorange2"
ITGA9	integrin subunit alpha 9	Adhesion	"darkorange2"	NOTCH4	notch 4	NOTCH signaling	"darkorange2"
ITGB4	integrin subunit beta 4	Adhesion	"darkorange2"	NRARP	NOTCH-regulated ankyrin repeat protein	NOTCH signaling	"darkorange2"
JAM2	junctional adhesion molecule 2	Adhesion	"darkorange2"	PDGFB	platelet-derived growth factor beta polypeptide	Growth factor	"darkorange2"
LAMA3	laminin subunit alpha 3	Adhesion	"darkorange2"	PTP4A3	protein tyrosine phosphatase type IVA, member 3	NOTCH signaling	"darkorange2"
LAMA4	laminin subunit alpha 4	Adhesion	"darkorange2"	VEGFC	vascular endothelial growth factor C	Growth factor	"darkorange2"
LAMA5	laminin subunit alpha 5	Adhesion	"darkorange2"	WISP2	WNT1 inducible signaling pathway protein 2	Wnt signaling	"darkorange2"
LAMB1	laminin subunit beta 1	Adhesion	"darkorange2"	WISP3	WNT1 inducible signaling pathway protein 3	Wnt signaling	"darkorange2"
LAMB1	laminin subunit beta 1	Adhesion	"darkorange2"	ZNF423	zinc finger protein 423	NOTCH signaling	"darkorange2"
MYH9	myosin heavy chain 9	Adhesion	"darkorange2"
PCDH1	protocadherin 1	Adhesion	"darkorange2"
PCDH12	protocadherin 12	Adhesion	"darkorange2"
PCDH17	protocadherin 17	Adhesion	"darkorange2"
PCDH19	protocadherin 19	Adhesion	"darkorange2"
TINAGL1	tubulointerstitial nephritis antigen like 1	Adhesion	"darkorange2"
ADAMTS7	ADAM metallopeptidase with thrombospondin type 1 motif, 7	ECM	"darkslateblue"
CNTN1	contactin 1	Adhesion	"darkslateblue"
COL5A3	collagen, type V, alpha 3	ECM	"darkslateblue"
COL6A1	collagen, type VI, alpha 1	ECM	"darkslateblue"
LAMA1	laminin, alpha 1	Adhesion	"darkslateblue"

The clusters “black”, “grey60”, and “lightyellow” show enrichment in genes associated with a proteoglycan rich extracellular matrix. Genes included in these clusters include the known NP markers type II collagen (COL2A1), type IX collagens (COL9A2, COL9A3), type XI collagen (COL11A2), aggrecan (ACAN), as well as other genes associated with a proteoglycan rich ECM that have not previously been associated with NP phenotype (Table 5). In contrast to the gene clusters previously discussed that were associated with synthesis of a fibrous matrix, these gene clusters express a diverse array of diffusible growth factors, with many being associated with the TGFβ signaling cascade. Associated growth factors include transforming growth factor alpha (TGFA), inhibin beta A (INHBA), inhibin alpha (INHA), growth differentiation factors (GDF5, GDF6), and bone morphogenetic proteins (BMP2, BMP6) and others (Table 5). The reliance on diffusible growth factors to mediate ECM homeostasis in a proteoglycan rich matrix such as that observed in the NP is logical since cells are usually separated by a thick matrix and have limited direct cell to cell contact. A comprehensive list of the genes associated with each regulatory cluster showing enrichment in either AF (fibrous) or NP (proteoglycan) markers are shown in Supplemental Table 4.

Table 5.

Nucleus pulposus co-regulatory gene network

ECM and cell adhesion related genes				Signaling associated genes
Gene symbol	Gene name	Function	Cluster	Gene symbol	Gene name	Function	Cluster
CDH26	cadherin-like 26	Adhesion	"black"	CTGF	connective tissue growth factor	Growth factor	"black"
CDHR5	mucin-like protocadherin	Adhesion	"black"	FGFR2	fibroblast growth factor receptor 2	Growth factor	"black"
CELSR3	cadherin, EGF LAG seven-pass G-type receptor 3	Adhesion	"black"	FGFR3	fibroblast growth factor receptor 3	Growth factor	"black"
CILP	cartilage intermediate layer protein, nucleotide pyrophosphohydrolase	ECM	"black"	BMP2	bone morphogenetic protein 2	Growth factor	"grey60"
COL27A1	collagen, type XXVII, alpha 1	ECM	"black"	BMP6	bone morphogenetic protein 6	Growth factor	"grey60"
COL2A1	collagen, type II, alpha 1	ECM	"black"	FGF1	fibroblast growth factor 1 (acidic)	Growth factor	"grey60"
CRTAP	cartilage associated protein	ECM	"black"	FGF2	fibroblast growth factor 2 (basic)	Growth factor	"grey60"
CTGF	connective tissue growth factor	ECM	"black"	GDF6	growth differentiation factor 6	Growth factor	"grey60"
DSCAML1	Down syndrome cell adhesion molecule like 1	Adhesion	"black"	HHIP	hedgehog interacting protein	Growth factor	"grey60"
GPC6	glypican 6	ECM	"black"	IGFBP3	insulin-like growth factor binding protein 3	Growth factor	"grey60"
ITGA10	integrin, alpha 10	Adhesion	"black"	INHBA	inhibin, beta A	Growth factor	"grey60"
LMLN	leishmanolysin-like	Adhesion	"black"	NGF	nerve growth factor (beta polypeptide)	Growth factor	"grey60"
PCDH20	protocadherin 20	Adhesion	"black"	NOG	noggin	Growth factor	"grey60"
TESK2	testis-specific kinase 2	Adhesion	"black"	PDGFC	platelet derived growth factor C	Growth factor	"grey60"
ADAMTS6	ADAM metallopeptidase with thrombospondin type 1 motif, 6	ECM	"grey60"	TGFA	transforming growth factor, alpha	Growth factor	"grey60"
CCDC80	coiled-coil domain containing 80	ECM	"grey60"	TGFBR1	transforming growth factor, beta receptor 1	Growth factor	"grey60"
CDH19	cadherin 19, type 2	Adhesion	"grey60"	TSHB	thyroid stimulating hormone, beta	Growth factor	"grey60"
CHI3L1	chitinase 3-like 1	ECM	"grey60"	VEGFA	vascular endothelial growth factor A	Growth factor	"grey60"
FN1	fibronectin 1	ECM	"grey60"	WNT1	wingless-type MMTV integration site family, member 1	Wnt signaling	"grey60"
HAPLN1	hyaluronan and proteoglycan link protein 1	ECM	"grey60"	WNT16	wingless-type MMTV integration site family, member 16	Wnt signaling	"grey60"
IMPG2	interphotoreceptor matrix proteoglycan 2	ECM	"grey60"	WNT9A	wingless-type MMTV integration site family, member 9A	Wnt signaling	"grey60"
ITGB5	integrin, beta 5	Adhesion	"grey60"	GDF5	growth differentiation factor 5	Growth factor	"lightyellow"
LAMB3	laminin, beta 3	Adhesion	"grey60"	HHIPL1	HHIP-like 1	Hedgehog signaling	"lightyellow"
LUM	lumican	ECM	"grey60"	HHIPL2	HHIP-like 2	Hedgehog signaling	"lightyellow"
PRG4	proteoglycan 4	ECM	"grey60"	INHA	inhibin, alpha	Growth factor	"lightyellow"
SERPINE1	serpin peptidase inhibitor, clade E member 1	ECM	"grey60"	NRG4	neuregulin 4	Growth factor	"lightyellow"
SERPINE2	serpin peptidase inhibitor, clade E member 2	ECM	"grey60"	NRTN	neurturin	Growth factor	"lightyellow"
SMOC1	SPARC related modular calcium binding 1	ECM	"grey60"
TIMP2	TIMP metallopeptidase inhibitor 2	ECM	"grey60"
TIMP3	TIMP metallopeptidase inhibitor 3	ECM	"grey60"
VCAN	versican	ECM	"grey60"
ACAN	aggrecan	ECM	"lightyellow"
ADAMTSL2	similar to ADAMTS-like 2; ADAMTS-like 2	ECM	"lightyellow"
BGN	biglycan	ECM	"lightyellow"
CHAD	chondroadherin	Adhesion	"lightyellow"
CILP2	cartilage intermediate layer protein 2	ECM	"lightyellow"
COL11A2	collagen, type XI, alpha 2	ECM	"lightyellow"
COL9A2	collagen, type IX, alpha 2	ECM	"lightyellow"
COL9A3	collagen, type IX, alpha 3	ECM	"lightyellow"
COMP	cartilage oligomeric matrix protein	ECM	"lightyellow"
EMILIN3	elastin microfibril interfacer 3	ECM	"lightyellow"
PRELP	proline/arginine-rich end leucine-rich repeat protein	ECM	"lightyellow"
SERPINA1	serpin peptidase inhibitor, clade A member 1	ECM	"lightyellow"
SPINT2	serine peptidase inhibitor, Kunitz type, 2	ECM	"lightyellow"

Discussion

The molecular phenotype of intervertebral spinal disk tissue, including the AF and NP, has been studied extensively over the past several years, primarily in animal models. The disk periphery is comprised of the fibrous annulus, derived from the scleroderm, while NP is derived from the notochord. However, notochordal cells in humans decrease in abundance with age, and are largely absent after adolescence^{38, 39}, although visible notochord tissue is present at maturity in other species. NP cells make predominantly type II collagen, whereas AF cells make both type I and type II collagen⁴⁰. The findings in our investigation utilizing high throughput RNA sequencing approaches are consistent with these findings in previous investigations, and also identify associations with other novel extracellular matrix proteins and associated regulatory factors.

Our initial clustering analysis performed using AF and NP specimens (Figure 1) demonstrates that blood content is an important consideration in the evaluation of surgically collected spinal disk tissues. Disk tissues have a very low density of cells, and the few cells that are present are usually encased in a thick extracellular matrix that makes RNA extraction technically challenging. Even the presence of small quantities of blood, from which RNA is much more easily extracted, can profoundly impact RNA content and resulting transcriptome data analyses if not carefully considered.

Our analysis reveals increased expression of known AF and NP markers within corresponding tissue types including enrichment of type I collagen in AF and a proteoglycan associated extracellular matrix enriched in genes such as ACAN, COMP, LUM, and others in the NP. We note that there is some overlap in mRNA expression between annulus and nucleus specimens. This overlap may reflect similarities in the developmental origin of these tissues or could be due to technical issues, for example, because there is some intermixing of annulus and nucleus cells during tissue harvest (e.g., in degenerative disk tissues with altered structural morphology).

These studies also implicate the WNT and NOTCH signaling pathway as a potentially important regulators of cell adhesion and matrix synthesis in AF tissue. These pathways are mediated by direct cell to cell interactions and have been shown to impact cellular adhesion and tissue integrity in various tissue types⁴¹. Golgi and ER related genes enriched in NP tissue may contribute to the production of the proteogylcan rich matrix associated with the NP environment. Therapeutic strategies that can increase protein output and upregulate the expression of NP specific genes have the potential to help disk tissue retain fluid and appropriate hydrostatic pressure, thus preventing disk space degeneration and associated disk space narrowing and osteoarthritis.

Recent studies have identified several novel AF and NP markers, our study shows support for many of these markers⁴². In our analyses, the proposed NP markers desmocollin 2 (DSC2)¹⁸, lubricin (PRG4)⁴³, and paired box 1 (PAX1)²⁰, showed co-regulation with networks enriched in NP related genes supporting their classification as NP markers. The novel AF markers brain abundant membrane attached signal protein 1 (BASP1), sclerostin domain containing 1 (SOSTDC1)¹⁸, glypican 3 (GPC3), and pleiotrophin (PTN)⁴⁴ also showed co-regulation with AF related ECM gene networks. Our study did not show a clear link to either AF or NP phenotypes for several published markers including CD24 antigen (CD24), keratin 8 (KRT8), keratin 18 (KRT18), keratin 19 (KRT19), cadherin 2 (CDH2)¹⁷, carbonic anhydrase 12 (CA12)⁴⁵, and hypoxia inducible factor 1 alpha subunit (HIF1A)^{13, 14, 46}, all of which showed co-regulation with gene networks unrelated to disk phenotype. Protein levels do not always correlate with mRNA expression, which could explain some of the differences between our study and previous investigations. Discrepancies could also be related to interspecies differences, as many of these published studies were carried out using non-human tissues. In addition, our study focused on evaluation of degenerative disc tissue, and it is possible that many of these markers may be present during early disk development and are gradually lost over time with aging and degeneration.

It is important to note that the gene relationships defined by network analyses in this study may exclude important functional/regulatory genes when a gene has a stronger relationship to another network. This was observed for the known AF related gene type I collagen (COL1 A1, COL1A2), which showed stronger co-regulation with bone related genes (gene cluster “royalblue”) rather than AF related genes. Despite this limitation, we were still able to identify large gene regulatory networks associated with ECM production in AF and NP tissues. Our analysis also does not take in account the numerous regulatory mechanisms that act in coordination with transcriptional mechanisms including protein phosphorylation and acetylation, histone modifications, microRNAs, and others. Future studies that integrate intervertebral disk transcriptomic profiles with various types of molecular data including microRNA profiles, and mass spectroscopy data may further help to elucidate novel molecular pathways involved intervertebral disk homeostasis.

This investigation provides a comprehensive overview of mRNA expression in annulus fibrosus and nucleus pulposus intervertebral disk tissue, including extracellular matrix components. By applying computational analyses to our large dataset of human clinical specimens, we have been able to identify candidate gene regulatory networks that act in AF and NP tissues to regulate extracellular matrix synthesis, an important determinant of intervertebral disk integrity. The transcriptome data generated in this study also serves as an important reference data set and has the potential to help solve many biological questions related to disk tissues. For example, our data can be used to evaluate the efficacy of tissue engineering strategies for intervertebral disk development. The data can also be applied to optimize stem cell differentiation strategies for therapeutic disk regeneration, as a variety of stem cell therapies are just beginning to be investigated in new clinical trials. Information generated in this study can also potentially be applied to identify novel therapeutic targets to enhance extracellular matrix synthesis and restore the normal mechanical properties of intervertebral disk tissue.

Supplementary Material

Supplemental Figure 1. Supplemental Figure 1.

Unsupervised hierarchical clustering of spine samples was performed to detect sample outliers. In total 10 samples were found to cluster independent from the majority of samples. These 10 samples expressed high levels of blood related mRNAs compared with the remaining samples.

Supplemental Table 1. Supplemental Table 1.

Clinical data associated with each disk specimen

Supplemental Table 2. Supplemental Table 2.

Hemoglobin expression in sample outliers

Supplemental Table 3. Supplemental Table 3.

Differential gene expression between AF and NP

Supplemental Table 4. Supplemental Table 4.

WGCNA gene clusters and associated genes

Clinical Significance.

This investigation provides important data on extracellular matrix gene regulatory networks in disk tissues. This information can be used to optimize pharmacologic, stem cell, and tissue engineering strategies for regeneration of the intervertebral disk and the treatment of back pain.